ref: ccae5d99d24a74f60a07a48ef98149d893fb15aa
parent: e47811ef8f045352dfad6c00add79e57e04fc96b
author: James Zern <[email protected]>
date: Tue May 5 07:40:13 EDT 2015
fix and enable vp9_dc_128_predictor_16x16 widen the loads and stores to 128-bit. this was added, but not enabled in: 493a857 Add some sse2 code for intra prediction. Change-Id: I277d7db608a7db7d75cc0bde86f48fa66ad487e4
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -171,7 +171,7 @@
specialize qw/vp9_dc_left_predictor_16x16/;
add_proto qw/void vp9_dc_128_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_128_predictor_16x16/;
+specialize qw/vp9_dc_128_predictor_16x16/, "$sse2_x86inc";
add_proto qw/void vp9_d207_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vp9_d207_predictor_32x32/, "$ssse3_x86inc";
--- a/vp9/common/x86/vp9_intrapred_sse2.asm
+++ b/vp9/common/x86/vp9_intrapred_sse2.asm
@@ -263,7 +263,7 @@
REP_RET
-INIT_MMX sse
+INIT_XMM sse2
cglobal dc_128_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset
GET_GOT goffsetq
@@ -270,7 +270,7 @@
DEFINE_ARGS dst, stride, stride3, lines4
lea stride3q, [strideq*3]
mov lines4d, 4
- movq m0, [GLOBAL(dc_128)]
+ mova m0, [GLOBAL(dc_128)]
.loop:
mova [dstq ], m0
mova [dstq+strideq ], m0